Demographics

Everyone

Demographics for all included participants.

Demographics
Summary
N Age (years) Education (years) Sex (M/F/O) EHI
844 29.08 (6.03) 14.38 (2.48) 441/391/12 5.09 (79.37)



Race n
White 606
Black or African American 82
Multiple 76
Asian 69
American Indian or Alaska Native 5
Native Hawaiian or Other Pacific Islander 3
Other 3


Hispanic ethnicity n
No 744
Yes 100


By handedness group

Demographics for included participants, by handedness group (EHI bins).

Handedness N Age (years) Education (years) Sex (M/F/O) EHI
Left 331 28.84 (6.1) 14.45 (2.39) 170/157/4 -81.61 (19.27)
Mixed 135 28.83 (6.17) 14.58 (2.6) 77/56/2 -8.89 (26.49)
Right 378 29.38 (5.93) 14.24 (2.5) 194/178/6 86.01 (16.61)
Left: (EHI <= -40) | Mixed: (-40 < EHI < 40) | Right: (EHI >= 40)


Field x Level x Handedness (binned)

Do we find an interaction of field x level x handedness, when handedness is binned as left (EHI <= -40) or right (EHI > +40)?

Summary. For reaction time, we find the critical interaction in the predicted direction (11.67ms, 95%CI [0.65, 22.69], p = .019, one-sided). Accuracy stats are in progress: the interaction effect will be close to zero, opposite the predicted direction (the point estimates are 1.76 for righties and 1.96 for lefties).

Reaction time

Plots

Error bars show 95% CI.






Statistics

Simple mixed regression model

Reaction time is modeled as a linear effect of field, level, and handedness, using data from every target-present trial with a “go” response:

lmer( rt ~ field*level*handedness + (1 | subject) )


Field by level by handedness interaction (RT)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df p.value1
9 1,166,282.858 1,166,367.139 −583,132.429 1,166,264.858 - - -
10 1,166,280.551 1,166,374.197 −583,130.276 1,166,260.551 4.307 1 .038
1 F-test (two-sided? https://daniellakens.blogspot.com/2016/04/one-sided-f-tests-and-halving-p-values.html)


Field by level by handedness interaction (RT)
Compare effect estimate to zero with emmeans()
field_consec level_consec handedness_consec estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF - RVF Local - Global Right - Left 11.666 5.622 Inf 0.648 22.685 2.075 .038
1 A positive number means LVF global bias is stronger in right handers (as predicted by AAH)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided


LVF Global bias by handedness bin (RT)
field_consec level_consec handedness estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF - RVF Local - Global Left 15.641 4.096 Inf 7.613 23.669 3.819 .0001
LVF - RVF Local - Global Mixed 21.658 6.414 Inf 9.087 34.228 3.377 .0007
LVF - RVF Local - Global Right 27.307 3.829 Inf 19.802 34.812 7.131 <.0001
1 A positive number means global bias (faster RT for global)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided, uncorrected


Field by level interaction (RT)
Old-school Omnibus F-test
term df sumsq meansq statistic p.value
field 1 1,664,691.722 1,664,691.722 23.612 <.0001
level 1 9,626,122.373 9,626,122.373 136.54 <.0001
handedness 1 10,185,712.46 10,185,712.46 144.477 <.0001
field:level 1 2,730,949.837 2,730,949.837 38.737 <.0001
field:handedness 1 1,505,316.949 1,505,316.949 21.352 <.0001
level:handedness 1 12,247.994 12,247.994 0.174 .677
field:level:handedness 1 127,954.685 127,954.685 1.815 .178
Residuals 86,205 6,077,502,195.866 70,500.576 - -


Accuracy

Plots

Error bars show 95% CI.




Statistics

Simple mixed regression model

In progress.

Accuracy is modeled as a binomial effect of field, level, and handedness, using binary correct/incorrect data from every target-present trial:

glmer( correct ~ field*level*handedness + (1 | subject), family = "binomial" )


Field by level by handedness interaction (Accuracy)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df p.value1
8 32,836.316 32,911.643 −16,410.158 32,820.316 - - -
9 32,837.667 32,922.41 −16,409.833 32,819.667 0.65 1 .42
1 F-test (two-sided? https://daniellakens.blogspot.com/2016/04/one-sided-f-tests-and-halving-p-values.html)


Field by level by handedness interaction (Accuracy)
Compare effect estimate to zero with emmeans()
field_consec level_consec handedness_consec odds.ratio1 SE df2 asymp.LCL3 asymp.UCL3 null z.ratio p.value4
LVF / RVF Local / Global Right / Left 1.113 0.145 Inf 0.862 1.436 1 0.82 .412
1 Backtransformed to odds ratio from log odds ratio (tests are performed on log odds ratio scale). A ratio > 1 means global bias is stronger in the LVF for right handers (predicted by AAH)
2 'Inf' df is expected when emmeans does logistic regression. See emmeans FAQ: https://cran.r-project.org/web/packages/emmeans/vignettes/FAQs.html#asymp.
3 Confidence level: 95%
4 Two-sided


LVF Global bias by handedness bin (Accuracy)
Compare effect estimate to zero with emmeans()
field_consec level_consec handedness odds.ratio1 SE df2 asymp.LCL3 asymp.UCL3 null z.ratio p.value4
LVF / RVF Local / Global Left 0.503 0.048 Inf 0.417 0.607 1 −7.193 <.0001
LVF / RVF Local / Global Mixed 0.916 0.136 Inf 0.685 1.225 1 −0.591 .554
LVF / RVF Local / Global Right 0.571 0.05 Inf 0.481 0.677 1 −6.414 <.0001
1 Backtransformed to odds ratio from log odds ratio (tests are performed on log odds ratio scale). A ratio > 1 means global bias is stronger in the LVF, as predicted for right handers. Mixed handers' global bias is shown here, but their data was not included in the binomial model.
2 'Inf' df is expected when emmeans does logistic regression. See emmeans FAQ: https://cran.r-project.org/web/packages/emmeans/vignettes/FAQs.html#asymp.
3 Confidence level: 95%
4 Two-sided


Field x Level x Handedness (continuous)

Do we find an interaction of field x level x handedness (continuous EHI score)?

Summary. For reaction time, we find the critical interaction in the predicted direction (.067ms per EHI unit, 95% CI [0.003, 0.13], p = .020, one-sided). This slope corresponds to 14.82ms for EHI = -100, and 28.14ms for EHI = +100. Accuracy stats are in progress.

Reaction time

Plots

Statistics

Model RT as a linear effect of field, level, and EHI (continuous):

rt_model_ehi <- lmer( rt ~ field*level*ehi + (1 | subject) )

Field by level by ehi interaction (RT)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df p.value1
9 1,387,640.958 1,387,726.807 −693,811.479 1,387,622.958 - - -
10 1,387,638.71 1,387,734.097 −693,809.355 1,387,618.71 4.248 1 .039
1 F-test (two-sided? https://daniellakens.blogspot.com/2016/04/one-sided-f-tests-and-halving-p-values.html)


Field by level interaction (RT)
Compare effect estimate to zero with emmeans()
field_consec level_consec estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
RVF - LVF Global - Local 0.067 0.032 Inf 0.003 0.13 2.061 .039
1 A positive number means global bias is stronger in LVF for right handers (as predicted), in ms per EHI unit (-100 to 100). Multiply this value by 200 to get the estimated difference in LVF global bias for strong left vs. right handers.
2 Z-approximation
3 Confidence level: 95%. Lower CI for one-sided test: 0.013.
4 Two-sided


Field by level interaction (RT)
Compare effect estimate to zero with emmeans()
contrast estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF Local - LVF Global 0.033 0.023 Inf −0.025 0.092 1.459 .463
RVF Local - RVF Global −0.033 0.023 Inf −0.092 0.026 −1.454 .465
1 A positive number means more global bias for right handers, in ms per EHI unit (-100 to 100)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided


Slope of EHI and RT by field and level
Compare effect estimate to zero with emmeans()
field level ehi.trend1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF Local 0.219 0.075 Inf 0.073 0.366 2.93 .003
RVF Local 0.096 0.075 Inf −0.05 0.243 1.288 .198
LVF Global 0.186 0.075 Inf 0.039 0.332 2.487 .013
RVF Global 0.13 0.075 Inf −0.017 0.276 1.734 .083
1 A positive number means slower RTs for right handers, in ms per EHI unit (-100 to 100).
2 Z-approximation
3 Confidence level: 95%
4 Two-sided




Estimated global bias by field, for EHI of -100 (strong left hander)
contrast estimate1 SE df z.ratio p.value
(LVF Local ehi-100) - (LVF Global ehi-100) 29.412 3.009 Inf 9.776 <.0001
(RVF Local ehi-100) - (RVF Global ehi-100) 14.59 3.016 Inf 4.837 <.0001
1 Estimated global bias (ms)


Estimated LVF Global Bias for EHI of -100 (strong left hander)
LVF_global_bias
14.822


Estimated global bias by field, for EHI of +100 (strong right hander)
contrast estimate1 SE df z.ratio p.value
LVF Local ehi100 - LVF Global ehi100 36.074 2.824 Inf 12.775 <.0001
RVF Local ehi100 - RVF Global ehi100 7.93 2.83 Inf 2.802 .026
1 Estimated global bias (ms)


Estimated LVF Global Bias for EHI of +100 (strong right hander)
LVF_global_bias
28.144


\[ 28.144 - 14.822 = 13.322ms \\ 13.322/200 = 0.067ms / EHI unit \] The model estimates that an average strong right hander (EHI +100) will have 13.32ms more LVF global bias than a strong left hander (EHI -100). Each unit change in EHI (-100:100) corresponds to a 0.067ms difference in LVF global bias. This is also the slope estimate given by the summary function:

summary(rt_model_ehi)
## Linear mixed model fit by REML ['lmerMod']
## Formula: rt ~ field:level:ehi + field:level + field:ehi + level:ehi +  
##     field + level + ehi + (1 | subject)
##    Data: aah_for_rt_ehi_model
## 
## REML criterion at convergence: 1387626.2
## 
## Scaled residuals: 
##       Min        1Q    Median        3Q       Max 
## -4.879612 -0.590631 -0.167132  0.363313  7.653749 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev.
##  subject  (Intercept) 28269.4  168.135 
##  Residual             42128.7  205.253 
## Number of obs: 102615, groups:  subject, 844
## 
## Fixed effects:
##                             Estimate  Std. Error   t value
## (Intercept)              680.1697334   5.9429525 114.44980
## fieldRVF                  -2.6755865   1.8307862  -1.46144
## levelGlobal              -32.7431087   1.8162558 -18.02781
## ehi                        0.2190687   0.0747656   2.93007
## fieldRVF:levelGlobal      21.4828945   2.5694569   8.36087
## fieldRVF:ehi              -0.1228165   0.0230212  -5.33493
## levelGlobal:ehi           -0.0333102   0.0228331  -1.45886
## fieldRVF:levelGlobal:ehi   0.0666075   0.0323174   2.06104
## 
## Correlation of Fixed Effects:
##             (Intr) fldRVF lvlGlb ehi    flRVF:G flRVF: lvlGl:
## fieldRVF    -0.155                                           
## levelGlobal -0.156  0.506                                    
## ehi         -0.064  0.010  0.010                             
## fldRVF:lvlG  0.110 -0.713 -0.706 -0.007                      
## fieldRVF:eh  0.010 -0.067 -0.033 -0.154  0.047               
## levelGlbl:h  0.010 -0.033 -0.065 -0.156  0.046   0.506       
## fldRVF:lvG: -0.007  0.047  0.046  0.110 -0.065  -0.712 -0.706




Test for a simple correlation between each subject’s EHI and LVF global bias.

Subject-level correlation: linear model
term estimate std.error statistic p.value1
(Intercept) 16.696 2.454 6.804 <.0001
ehi 0.042 0.031 1.374 .17
1 Two-sided


Subject-level correlation: Spearman's rho
rho statistic p.value1 method alternative
0.041 96,127,760.619 .119 Spearman's rank correlation rho greater
1 One-sided



Accuracy

Plots

Statistics

In progress. Model accuracy as a binomial effect of field, level, and EHI (continuous):

acc_ehi_model <- glmer( rt ~ field*level*ehi + (1 | subject), family = "binomial" )


Field by level by EHI interaction (Accuracy)
ANOVA: compare models with vs. without interaction term
npar AIC BIC logLik deviance Chisq Df p.value1
8 39,111.468 39,188.189 −19,547.734 39,095.468 - - -
9 39,113.409 39,199.72 −19,547.704 39,095.409 0.059 1 .808
1 F-test (two-sided? https://daniellakens.blogspot.com/2016/04/one-sided-f-tests-and-halving-p-values.html)


Field by level interaction (acc)
Compare effect estimate to zero with emmeans()
field_consec level_consec estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
RVF - LVF Local - Global 0 0.001 Inf −0.001 0.002 0.249 .803
1 A positive number means global bias is stronger in LVF for right handers (as predicted), in logodds per EHI unit (-100 to 100). Multiply this value by 200 to get the estimated difference in LVF global bias for strong left vs. right handers.
2 Z-approximation
3 Confidence level: 95%. Lower CI for one-sided test: 0.013.
4 Two-sided


Field by level interaction (acc)
Compare effect estimate to zero with emmeans()
contrast estimate1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF Global - LVF Local −0.001 0.001 Inf −0.002 0 −1.8 .072
RVF Global - RVF Local −0.001 0.001 Inf −0.002 0 −2.359 .018
1 A positive number means more global bias, in logodds per EHI unit (-100 to 100)
2 Z-approximation
3 Confidence level: 95%
4 Two-sided, uncorrected


Slope of EHI and acc by field and level
Compare effect estimate to zero with emmeans()
field level ehi.trend1 SE df2 asymp.LCL3 asymp.UCL3 z.ratio p.value4
LVF Global −0.001 0.001 Inf −0.002 0.001 −1.026 .305
RVF Global −0.001 0.001 Inf −0.002 0.001 −0.956 .339
LVF Local 0 0.001 Inf −0.001 0.001 0.579 .562
RVF Local 0.001 0.001 Inf −0.001 0.002 1.067 .286
1 A positive number means higher accuracy for right handers, in logodds per EHI unit (-100 to 100).
2 Z-approximation
3 Confidence level: 95%
4 Two-sided

Generalized linear mixed model fit by maximum likelihood (Laplace Approximation) [glmerMod] Family: binomial ( logit ) Formula: correct ~ field * level * ehi + (1 | subject) Data: aah_for_acc_ehi_model

 AIC      BIC   logLik deviance df.resid 

39113.4 39199.7 -19547.7 39095.4 108023

Scaled residuals: Min 1Q Median 3Q Max -11.748088 0.117364 0.168282 0.238851 1.067892

Random effects: Groups Name Variance Std.Dev. subject (Intercept) 1.05841 1.02879 Number of obs: 108032, groups: subject, 844

Fixed effects: Estimate Std. Error z value Pr(>|z|)
(Intercept) 3.9930184017 0.0532670291 74.96229 < 2e-16 fieldRVF -0.4209327473 0.0474500751 -8.87107 < 2e-16 levelLocal -0.9403339363 0.0440725615 -21.33604 < 2e-16 ehi -0.0006744722 0.0006575342 -1.02576 0.305005
fieldRVF:levelLocal 0.5311871060 0.0594032015 8.94206 < 2e-16
fieldRVF:ehi 0.0000955515 0.0005978575 0.15982 0.873020
levelLocal:ehi 0.0009994867 0.0005551631 1.80035 0.071806 .
fieldRVF:levelLocal:ehi 0.0001864551 0.0007484763 0.24911 0.803273
— Signif. codes: 0 ‘’ 0.001 ’’ 0.01 ’’ 0.05 ‘.’ 0.1 ’ ’ 1

Correlation of Fixed Effects: (Intr) fldRVF lvlLcl ehi flRVF:L flRVF: lvlLc: fieldRVF -0.528
levelLocal -0.575 0.636
ehi -0.084 0.058 0.064
fldRVF:lvlL 0.423 -0.799 -0.742 -0.046
fieldRVF:eh 0.057 -0.105 -0.069 -0.538 0.084
levelLocl:h 0.063 -0.069 -0.090 -0.583 0.066 0.635
fldRVF:lvL: -0.045 0.084 0.066 0.430 -0.081 -0.799 -0.741 optimizer (Nelder_Mead) convergence code: 0 (OK) Model failed to converge with max|grad| = 0.00445801 (tol = 0.002, component 1) Model is nearly unidentifiable: very large eigenvalue - Rescale variables?

Estimated global bias by field, for EHI of -100 (strong left hander)
contrast odds.ratio1 SE df null z.ratio p.value
(LVF Global ehi-100) / (LVF Local ehi-100) 2.83 0.209 Inf 1 14.075 <.0001
(RVF Global ehi-100) / (RVF Local ehi-100) 1.695 0.112 Inf 1 7.953 <.0001
1 Estimated global bias. An odds ratio > 1 means global bias.


Estimated LVF Global Bias for EHI of -100 (strong left hander)
LVF_global_bias
1.135


Estimated global bias by field, for EHI of +100 (strong right hander)
contrast odds.ratio1 SE df null z.ratio p.value
LVF Global ehi100 / LVF Local ehi100 2.317 0.157 Inf 1 12.409 <.0001
RVF Global ehi100 / RVF Local ehi100 1.337 0.083 Inf 1 4.697 <.0001
1 Estimated global bias (ms)


Estimated LVF Global Bias for EHI of +100 (strong right hander)
LVF_global_bias
0.98

\[ log(.98) = -0.0202 log(0.135) == 0.1266 -0.0202 - 0.1266 = -0.147 (logodds) -0.147 / 200 = -0.00007 logodds / EHI unit \] The model estimates that an average strong left hander (EHI -100) will have 1.135/0.98 greater relative odds of correctness for LVF global stimuli versus a strong right hander (EHI +100). Each unit change in EHI (-100:100) corresponds to a 0.00007 (logodds) difference in LVF global bias. This is close to the slope estimate given by the summary function (the difference should be due to rounding:

summary(acc_model_ehi)
## Generalized linear mixed model fit by maximum likelihood (Laplace
##   Approximation) [glmerMod]
##  Family: binomial  ( logit )
## Formula: correct ~ field * level * ehi + (1 | subject)
##    Data: aah_for_acc_ehi_model
## 
##      AIC      BIC   logLik deviance df.resid 
##  39113.4  39199.7 -19547.7  39095.4   108023 
## 
## Scaled residuals: 
##        Min         1Q     Median         3Q        Max 
## -11.748088   0.117364   0.168282   0.238851   1.067892 
## 
## Random effects:
##  Groups  Name        Variance Std.Dev.
##  subject (Intercept) 1.05841  1.02879 
## Number of obs: 108032, groups:  subject, 844
## 
## Fixed effects:
##                              Estimate    Std. Error   z value Pr(>|z|)    
## (Intercept)              3.9930184017  0.0532670291  74.96229  < 2e-16 ***
## fieldRVF                -0.4209327473  0.0474500751  -8.87107  < 2e-16 ***
## levelLocal              -0.9403339363  0.0440725615 -21.33604  < 2e-16 ***
## ehi                     -0.0006744722  0.0006575342  -1.02576 0.305005    
## fieldRVF:levelLocal      0.5311871060  0.0594032015   8.94206  < 2e-16 ***
## fieldRVF:ehi             0.0000955515  0.0005978575   0.15982 0.873020    
## levelLocal:ehi           0.0009994867  0.0005551631   1.80035 0.071806 .  
## fieldRVF:levelLocal:ehi  0.0001864551  0.0007484763   0.24911 0.803273    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr) fldRVF lvlLcl ehi    flRVF:L flRVF: lvlLc:
## fieldRVF    -0.528                                           
## levelLocal  -0.575  0.636                                    
## ehi         -0.084  0.058  0.064                             
## fldRVF:lvlL  0.423 -0.799 -0.742 -0.046                      
## fieldRVF:eh  0.057 -0.105 -0.069 -0.538  0.084               
## levelLocl:h  0.063 -0.069 -0.090 -0.583  0.066   0.635       
## fldRVF:lvL: -0.045  0.084  0.066  0.430 -0.081  -0.799 -0.741
## optimizer (Nelder_Mead) convergence code: 0 (OK)
## Model failed to converge with max|grad| = 0.00445801 (tol = 0.002, component 1)
## Model is nearly unidentifiable: very large eigenvalue
##  - Rescale variables?



Test for a simple correlation between each subject’s EHI and LVF global bias.

Subject-level correlation: linear model
term estimate std.error statistic p.value1
(Intercept) 1.987 0.311 6.381 <.0001
ehi 0.002 0.004 0.448 .655
1 Two-sided


Subject-level correlation: Spearman's rho
rho statistic p.value1 method alternative
0.036 96,571,310.453 .147 Spearman's rank correlation rho greater
1 One-sided